code</p>\n","updatedAt":"2026-08-20T06:51:50.094Z","author":{"_id":"62b3a4cf003cd12329e0a822","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62b3a4cf003cd12329e0a822/nZTj3yNcoYlQ2ESCREM0l.jpeg","fullname":"Igor Itkin","name":"BukaByaka","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.33263149857521057},"editors":["BukaByaka"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/62b3a4cf003cd12329e0a822/nZTj3yNcoYlQ2ESCREM0l.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.18115","authors":[{"_id":"6a86a3cfdb13816030683fb4","name":"Igor Itkin","hidden":false}],"publishedAt":"2026-06-30T00:00:00.000Z","submittedOnDailyAt":"2026-08-20T00:00:00.000Z","title":"Temporal Multi-Signal Fusion for Token-Level Hallucination Detection","submittedOnDailyBy":{"_id":"62b3a4cf003cd12329e0a822","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62b3a4cf003cd12329e0a822/nZTj3yNcoYlQ2ESCREM0l.jpeg","isPro":false,"fullname":"Igor Itkin","user":"BukaByaka","type":"user","name":"BukaByaka"},"summary":"Token-level hallucination detectors score each token independently from a single signal, and fail exactly when the generating model is confidently wrong. This paper instead treats hallucination as a temporally extended span and detects it by sequence labeling: each token is scored from a 33-dimensional feature stream that fuses text statistics, Natural Language Inference (NLI) entailment, and language model surprisal, with no access to model internals. A Bidirectional Gated Recurrent Unit (BiGRU) over these features reaches an AUC of 0.840 on RAGTruth (10 seeds), an 11-point gain over an independent logistic-regression baseline (p = 0.002, Wilcoxon signed-rank). A controlled decomposition attributes most of the gain to temporal order rather than model capacity: evidence propagates from confident positions to ambiguous neighbors within a span. The same 0.845 ceiling recurs across recurrent, state-space (Mamba), and attention architectures, locating the bottleneck in the feature set rather than the model. Because it reads only the generated text and external signals, the detector works on closed-source models, and it keeps working on text produced by language models it never saw during training, losing under 4% AUC.","upvotes":2,"discussionId":"6a86a3d0db13816030683fb5","githubRepo":"https://github.com/YehudaItkin/temporal-hallucination-detection","githubRepoAddedBy":"user","ai_summary":"Hallucination is detected as temporally extended spans via sequence labeling over fused external features, achieving robust cross-model performance without internal model access.","ai_keywords":["sequence labeling","BiGRU","Natural Language Inference","language model surprisal","RAGTruth","Mamba","state-space models","cross-model generalization"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":0},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"62b3a4cf003cd12329e0a822","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62b3a4cf003cd12329e0a822/nZTj3yNcoYlQ2ESCREM0l.jpeg","isPro":false,"fullname":"Igor Itkin","user":"BukaByaka","type":"user"},{"_id":"63ac5701c21e60a3e9b58aa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63ac5701c21e60a3e9b58aa7/g6EX7diOpuA94R2ab-rZC.png","isPro":true,"fullname":"Dipankar Sarkar","user":"dipankarsarkar","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.18115.md","query":{}}">
Temporal Multi-Signal Fusion for Token-Level Hallucination Detection
Abstract
Hallucination is detected as temporally extended spans via sequence labeling over fused external features, achieving robust cross-model performance without internal model access.
Token-level hallucination detectors score each token independently from a single signal, and fail exactly when the generating model is confidently wrong. This paper instead treats hallucination as a temporally extended span and detects it by sequence labeling: each token is scored from a 33-dimensional feature stream that fuses text statistics, Natural Language Inference (NLI) entailment, and language model surprisal, with no access to model internals. A Bidirectional Gated Recurrent Unit (BiGRU) over these features reaches an AUC of 0.840 on RAGTruth (10 seeds), an 11-point gain over an independent logistic-regression baseline (p = 0.002, Wilcoxon signed-rank). A controlled decomposition attributes most of the gain to temporal order rather than model capacity: evidence propagates from confident positions to ambiguous neighbors within a span. The same 0.845 ceiling recurs across recurrent, state-space (Mamba), and attention architectures, locating the bottleneck in the feature set rather than the model. Because it reads only the generated text and external signals, the detector works on closed-source models, and it keeps working on text produced by language models it never saw during training, losing under 4% AUC.
Community
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2608.18115 in a model README.md to link it from this page.
Cite arxiv.org/abs/2608.18115 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2608.18115 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.